|
import gradio as gr |
|
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline |
|
import torch |
|
|
|
|
|
checkpoint = "microsoft/phi-2" |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(checkpoint, trust_remote_code=True) |
|
model = AutoModelForCausalLM.from_pretrained(checkpoint, torch_dtype=torch.float32, device_map="cpu", trust_remote_code=True) |
|
|
|
|
|
phi2 = pipeline("text-generation", tokenizer=tokenizer, model=model) |
|
|
|
|
|
def generate(prompt, chat_history): |
|
|
|
instruction = "You are a helpful assistant to 'User'. You do not respond as 'User' or pretend to be 'User'. You only respond once as 'Assistant'." |
|
final_prompt = f"Instruction: {instruction}\n" |
|
|
|
for sent, received in chat_history: |
|
final_prompt += "User: " + sent + "\n" |
|
final_prompt += "Assistant: " + received + "\n" |
|
|
|
final_prompt += "User: " + prompt + "\n" |
|
final_prompt += "Output:" |
|
|
|
generated_text = phi2(final_prompt, max_new_tokens=16)[0]["generated_text"] |
|
response = generated_text.split("Output:")[1].split("User:")[0] |
|
|
|
if "Assistant:" in response: |
|
response = response.split("Assistant:")[1].strip() |
|
|
|
chat_history.append((prompt, response)) |
|
|
|
return "", chat_history |
|
|
|
|
|
with gr.Blocks() as demo: |
|
gr.Markdown("# Phi-2 Chatbot Demo") |
|
|
|
chatbot = gr.Chatbot() |
|
msg = gr.Textbox() |
|
|
|
clear = gr.ClearButton([msg, chatbot]) |
|
msg.submit(fn=generate, inputs=[msg, chatbot], outputs=[msg, chatbot]) |
|
|
|
demo.launch() |